Jul 28, 2025, 3:15 PM ETInterview, Fireside Chat, Podcast

The Thermodynamics of Trading with Daniel Pontecorvo

Jane Street's physical engineering team manages the lifecycle of all physical spaces, including leasing, design, construction, and operation of offices, data centers, and co-location sites.
Proximity is a critical physical constraint for trading firms; latency concerns extend to fiber length measured in microseconds.
Data centers must balance scalability with fixed location constraints; running out of space or power in a specific site creates distance issues if a new site is required.
Cooling is the second-largest power consumer in a data center, following IT equipment itself.
Power Utilization Efficiency (PUE) measures total power consumption versus compute power, ranging from efficient 1.1 (10% waste) to inefficient 1.8 or 2.0.
Efficiency gains are achieved by utilizing economizer cycles (using outside air) to reduce reliance on mechanical cooling compressors.

Chilled Water Systems: Basic data centers use water cooled to 50–65°F, circulated via air-cooled chillers and Cray units (heat exchangers) to absorb heat from server air.
Delta T Optimization: The industry targets a 20–30°F temperature differential (Delta T) to balance manageable airflow rates with effective heat rejection.
Rear Door Heat Exchangers: Bolted directly to the back of server racks, these units reduce hot air recirculation risks and lower fan energy consumption, which scales with the cube of air velocity.
Containment Strategies: Physical ductwork separates "cold aisles" (air intake) and "hot aisles" (air exhaust) to prevent short-cycling mixing, which creates localized hot spots.
Failure Scenario Planning: Running servers at higher ambient temperatures reduces efficiency but increases risk; redundant mechanical equipment and early failure communication protocols are essential.

Custom Monitoring Stack: Jane Street developed proprietary software to aggregate data from disparate power, cooling, and lighting systems into a single pane of glass, allowing custom alert thresholds.
Alerting Strategy: The team manages a "Goldilocks problem," tuning alerts to avoid false positives (alert fatigue) while ensuring early detection of genuine issues.
Incident Case Study: A maintenance event on a chilled water system led to a drained water supply during trading hours; the custom monitoring system detected temperature rises at the rack level faster than the provider, enabling a proactive business response.
Fire Suppression Design: Data centers utilize pre-action systems where water is excluded from pipes until heat/smoke is detected, preventing accidental water discharge from standard temperature melt heads.
High-Temperature Sprinklers: The team upgraded to higher temperature-rated sprinkler heads to prevent melting during temporary cooling failures that are not true fires.
On-Site Response: The physical engineering team maintains a dedicated on-site response team to supervise recovery and calculate refill times (e.g., flow rates into 12-inch pipes) to report accurate recovery windows to traders.

Density Increase: Traditional data center racks handled 10–15 kW; modern GPU setups now require designs up to 170 kW per rack, with future projections of 600 kW.
Liquid Cooling Necessity: Water is 3,000 to 4,000 times more effective at heat transfer than air due to higher specific heat capacity and density.
Cooling Architectures:
- Rear Door Heat Exchangers: Water is brought inches from the rack without touching components.
- Direct Liquid Cooling (DLC): Cold plates with micro-channels are bolted directly to GPUs/chips using thermal paste; requires high-purity water with filtration and anti-corrosion additives.
- Immersion Cooling: Servers are dunked in dielectric fluid, though this presents the highest risk profile regarding leaks and safety.
Leak Mitigation: Physical design minimizes mechanical joints (preferring off-site welding) and installs leak detection specifically at connection points; fluid quality control prevents clogging and corrosion.
Power Distribution Constraints: High densities create "power hotspots" where a single device consumes the capacity previously allocated to multiple racks, requiring complex UPS clustering and DC/AC power distribution strategies.
Networking Physicality: Modern GPU clusters require "rail-optimized networks" with direct GPU-to-GPU wiring, leading to massive cable density within racks that competes with space for power and cooling infrastructure.

Modular Desk Ecosystem: All desks are mounted on locking wheels and utilize a standardized global design to facilitate weekly team reconfigurations.
Underfloor Cooling: A raised floor (12–16 inches) creates a pressurized plenum that supplies 60–65°F air directly to the breathing zone and under-desk PC areas; diffusers allow individual temperature adjustment.
Network Infrastructure: The office uses fiber to the end of the row with "end-of-row" switches to avoid the airflow blockage and signal limitations of long copper runs found in "top-of-rack" designs.
Circadian Lighting: Lighting systems adjust color temperature throughout the day (warm ~2,700K in morning, cool ~4,000K at midday) to align with human circadian rhythms.
CO2 Mitigation: Indoor CO2 levels are monitored, with goals to keep levels below 1,500 ppm to prevent cognitive decline; strategies include increasing outside air intake (energy trade-off) and testing CO2 scrubbers.
Cultural Impact: Weekly desk moves are engineered to optimize team adjacencies for collaboration, allowing new joiners to sit with tenured staff and shifting teams based on evolving project needs.
Air Filtration: The system moves double the normal air volume through MERV 16 (hospital-grade) filters to maintain fresh air quality at the breathing zone.

Role Definition: The physical engineering team is recruited for its ability to bridge mechanical/electrical engineering with high-stakes business impact.
Hiring Mix: Jane Street seeks a blend of industry veterans and recent graduates; the latter are valued for challenging status quo assumptions (e.g., moving desks, implementing DLC) without preconceived constraints.
Impact Visibility: The team emphasizes direct feedback loops with traders and developers to demonstrate how physical infrastructure decisions directly influence trading performance and culture.